Scuttlebutt

Scuttlebutt

Data centers in space don't pencil, but ocean buoys might, says Meta's former CTO

15 insights: Fusion that could power Austin for a year on a pickup truck of fuel, why he says the US grid will not break, and what Zuckerberg's founder mode looks like

Scuttlebutt's avatar
Scuttlebutt
Sep 25, 2026
∙ Paid

Power Players with Brian Sozzi · Sep 24, 2026 · 31m · 17m saved

With Mike Schroepfer, Founding Partner at Gigascale Capital, formerly CTO of Meta

Data centers in space don't pencil, but ocean buoys might, says Meta's former CTO

A band of the Southern Ocean holds 10 terawatts of wave energy that nobody harnesses, Mike Schroepfer said, and one of his companies wants to use it to answer AI queries. Its test buoys go in the water off the US West Coast this year.

Much of the technology industry is talking about putting data centers in space. Schroepfer, who backs the buoy company, said the ocean is far cheaper to reach, already cold enough to cool the chips, and far from any neighbor who objects.

"So it's a floating, autonomous AI inference data center."

Mike Schroepfer, founding partner of Gigascale Capital, on Power Players with Brian Sozzi. He spent 18 years at Meta, grew its engineering team from 150 people to 35,000, was CTO from 2013 to 2022, and led the deals for Oculus and Instagram. He still advises the company.

The full interview is covered here so you can skip it. 31 minutes of audio, 14 minutes of reading.

Here are the 15 insights that matter.


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • Wave-powered buoys could run AI inference at sea, with commercial operation maybe next year

  • Putting material in the ocean costs an order of magnitude less than lifting it into space

  • Fusion is a science success and an economics question, and whoever solves the economics has "a license to print money"

  • New gas plants built for AI will run for 20 to 30 years, which he called a debt society is taking on

  • The grid will not break, because AI models can be made 10x to 100x more efficient and US power generation can double

  • Space data centers are a good business for SpaceX and a hard one for everyone else

  • A Chinese model better than the best US models would be a Sputnik moment

  • The leading AI labs need a lead worth a premium to pay for training, which he called the $2 trillion question

  • Founder mode at Meta means a 30-page email on Sunday night from Mark Zuckerberg

1. A Buoy That Runs AI

Brian Sozzi opened with the autonomous buoys he had seen in a picture while researching Schroepfer.

10 terawatts of wave energy sits unused

Schroepfer said the question is where to get much more energy cheaply and reliably with little damage to the environment. A band of the Southern Ocean holds 10 terawatts of wave energy that is not harnessed at all. The company's machine is a buoy that bobs up and down and uses that motion to pump water through a turbine, the way water spins a turbine in a hydropower dam.

The power stays at sea and runs AI

The obvious question is what to do with the power. With Starlink providing bandwidth and cold ocean water solving cooling, he said, the buoy is the perfect platform for AI inference, the computing that answers users' queries.

"So it's a floating, autonomous AI inference data center." — Mike Schroepfer

Testing this year, selling maybe next year

Sozzi listed Schroepfer's Meta career and asked when the buoys would appear. Schroepfer said test units will go in off the US West Coast this year, with commercial operation next year, maybe.

"It's not years out, it's in the near future." — Mike Schroepfer

2. Ocean Beats Space on Cost

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture